def test_batch_apply_component_with_dict_input_space(self):
input_space = Dict(dict(a=FloatBox(shape=(3, )),
b=FloatBox(shape=(1, 2))),
add_batch_rank=True,
add_time_rank=True)
sub_component = DictPreprocessorStack(preprocessors=dict(
a=[Multiply(factor=2.0)],
b=[Divide(divisor=2.0), Multiply(factor=2.0)]))
batch_apply = BatchApply(sub_component=sub_component,
api_method_name="preprocess")
test = ComponentTest(component=batch_apply,
input_spaces=dict(input_=input_space))
sample = input_space.sample(size=(5, 10))
sample_folded_a = np.reshape(sample["a"], newshape=(50, 3))
sample_folded_b = np.reshape(sample["b"], newshape=(50, 1, 2))
expected_a = sample_folded_a * 2.0
expected_a = np.reshape(expected_a, newshape=(5, 10, 3))
expected_b = sample_folded_b
expected_b = np.reshape(expected_b, newshape=(5, 10, 1, 2))
test.test(("apply", sample),
expected_outputs=dict(a=expected_a, b=expected_b))
def test_batch_apply_component_with_simple_input_space(self):
input_space = FloatBox(shape=(3,), add_batch_rank=True, add_time_rank=True)
sub_component = DenseLayer(units=4, biases_spec=False)
batch_apply = BatchApply(sub_component=sub_component, api_method_name="call")
test = ComponentTest(component=batch_apply, input_spaces=dict(input_=input_space))
weights = test.read_variable_values(batch_apply.variable_registry["batch-apply/dense-layer/dense/kernel"])
sample = input_space.sample(size=(5, 10))
sample_folded = np.reshape(sample, newshape=(50, 3))
expected = dense_layer(sample_folded, weights)
expected = np.reshape(expected, newshape=(5, 10, 4))
test.test(("call", sample), expected_outputs=expected)